Skip to main content
Market value is what the secondary market is willing to pay right now. We read it from Jupiter’s routing engine — the same way an actual swap would execute.

​
The query

The probe size is configured per asset to be large enough that we sample real depth, but small enough that we don’t move the market. For deep assets like USDC ↔ USDT it’s higher; for thinner assets like Hylo’s hyloSOL+ it’s smaller. We always swap to USDC and then convert USDC → USD via Pyth. This keeps the numeraire consistent: every asset’s market_usd is denominated the same way, and USDC’s own market price (a depeg signal) doesn’t contaminate other assets’ spreads.

​
Why Jupiter and not a single DEX

Jupiter’s router aggregates depth across Orca, Raydium, Phoenix, Meteora and the LST-specialized AMMs. Reading a single DEX gives you that DEX’s idiosyncratic liquidity profile. Reading Jupiter’s route gives you the price an actual user would receive — which is what matters for the alert. The mSOL/Pyth case study is the canonical counter-example: a single thin pool (Mango) was the entire oracle input. The resulting flash crash liquidated $21M of healthy collateral. If the oracle had read a routed quote, the thin pool would have weighted accordingly and the print would not have happened.

​
Polling cadence

  • Jupiter market quote: a single round-robin loop ticks every jupiter_tick_ms (≈2.5s) and fires a small batch of quotes per tick. Across the ~19 Jupiter-backed slots this works out to ≈50s per asset — the cadence stays under the keyless lite-api RPS budget.
  • Sanctum LST intrinsic: a single batch call for all LSTs at once every 15s (sanctum_poll_ms).
We do not query Jupiter on every Pyth tick — that would melt the API quota and generate noise on micro-prints. The round-robin pacing keeps the market poll time-aligned with the intrinsic feeds the engine’s recompute consumes. The Pyth intrinsic stream reconnects on transport errors with exponential backoff (500ms → 30s) plus ±50% jitter — that jitter applies to the stream reconnect, not to the Jupiter market poll.

​
What we don’t model

Slippage at size. Pegana’s probe is fixed; if you intend to swap $5M of an asset, your effective market price will be worse than what we publish. The signal is calibrated for small-to-medium trades and for state monitoring. For large trades, use Jupiter’s quote API directly with your real size. Routing instability. A route may exist this second and not the next. We accept that as long as Jupiter returns some quote. If routing fails entirely (no path to USDC), the engine skips the recompute and holds the last published state until a route returns — it does not flip the asset to UNKNOWN. MEV-extracted user price. Our quote is what the route returns. A real swap may receive a slightly worse price because of MEV between submission and inclusion. We do not adjust for this — the gap is typically small enough not to matter for peg state, though it matters a lot for actual execution.

​
Caching and freshness

  • The router’s quote is cached for the polling interval (no double-call inside 15s)
  • The engine refuses to consume a quote older than 30s (staleness gate)
  • On a stale quote the engine skips the recompute and holds the last published state rather than emitting a fresh verdict on degraded input (it does not flip to UNKNOWN)

​
Next

The spread

The formula, EWMA smoothing, per-asset thresholds, the 5-state FSM.